TrueSight-ES: Machine Learning Approach to Detecting Splice Junctions in RNA-Seq Data

RNA-seq reads carry substantial information on gene organization and gene expression. The task of RNA-seq read alignment to a reference genome is challenging if the read is spanning a splice junction. Still, this task is critically important for reliable identification of novel splice junctions and alternative gene isoforms. To solve the gapped alignment problem existing programs use various heuristics, such as “exon inference” (TopHat) and “anchor extension” (MapSplice and others). A recently developed program called TrueSight uses sequence patterns in reference genome to make identification of true splice junctions more reliable. Here, we present a result of further development, the program called TrueSight-ES, in which parameters of the models of splice junctions are determined in iterations of unsupervised training by GeneMark-ES. The advantage of this approach is the improved splice junction prediction accuracy as well as the ability to work with arbitrarily small RNA-seq data sets.